1 /*
2 * gendisk handling
3 */
4
5 #include <linux/module.h>
6 #include <linux/fs.h>
7 #include <linux/genhd.h>
8 #include <linux/kdev_t.h>
9 #include <linux/kernel.h>
10 #include <linux/blkdev.h>
11 #include <linux/init.h>
12 #include <linux/spinlock.h>
13 #include <linux/proc_fs.h>
14 #include <linux/seq_file.h>
15 #include <linux/slab.h>
16 #include <linux/kmod.h>
17 #include <linux/kobj_map.h>
18 #include <linux/mutex.h>
19 #include <linux/idr.h>
20 #include <linux/log2.h>
21 #include <linux/pm_runtime.h>
22
23 #include "blk.h"
24
25 static DEFINE_MUTEX(block_class_lock);
26 struct kobject *block_depr;
27
28 /* for extended dynamic devt allocation, currently only one major is used */
29 #define NR_EXT_DEVT (1 << MINORBITS)
30
31 /* For extended devt allocation. ext_devt_lock prevents look up
32 * results from going away underneath its user.
33 */
34 static DEFINE_SPINLOCK(ext_devt_lock);
35 static DEFINE_IDR(ext_devt_idr);
36
37 static struct device_type disk_type;
38
39 static void disk_check_events(struct disk_events *ev,
40 unsigned int *clearing_ptr);
41 static void disk_alloc_events(struct gendisk *disk);
42 static void disk_add_events(struct gendisk *disk);
43 static void disk_del_events(struct gendisk *disk);
44 static void disk_release_events(struct gendisk *disk);
45
46 /**
47 * disk_get_part - get partition
48 * @disk: disk to look partition from
49 * @partno: partition number
50 *
51 * Look for partition @partno from @disk. If found, increment
52 * reference count and return it.
53 *
54 * CONTEXT:
55 * Don't care.
56 *
57 * RETURNS:
58 * Pointer to the found partition on success, NULL if not found.
59 */
disk_get_part(struct gendisk * disk,int partno)60 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
61 {
62 struct hd_struct *part = NULL;
63 struct disk_part_tbl *ptbl;
64
65 if (unlikely(partno < 0))
66 return NULL;
67
68 rcu_read_lock();
69
70 ptbl = rcu_dereference(disk->part_tbl);
71 if (likely(partno < ptbl->len)) {
72 part = rcu_dereference(ptbl->part[partno]);
73 if (part)
74 get_device(part_to_dev(part));
75 }
76
77 rcu_read_unlock();
78
79 return part;
80 }
81 EXPORT_SYMBOL_GPL(disk_get_part);
82
83 /**
84 * disk_part_iter_init - initialize partition iterator
85 * @piter: iterator to initialize
86 * @disk: disk to iterate over
87 * @flags: DISK_PITER_* flags
88 *
89 * Initialize @piter so that it iterates over partitions of @disk.
90 *
91 * CONTEXT:
92 * Don't care.
93 */
disk_part_iter_init(struct disk_part_iter * piter,struct gendisk * disk,unsigned int flags)94 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
95 unsigned int flags)
96 {
97 struct disk_part_tbl *ptbl;
98
99 rcu_read_lock();
100 ptbl = rcu_dereference(disk->part_tbl);
101
102 piter->disk = disk;
103 piter->part = NULL;
104
105 if (flags & DISK_PITER_REVERSE)
106 piter->idx = ptbl->len - 1;
107 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
108 piter->idx = 0;
109 else
110 piter->idx = 1;
111
112 piter->flags = flags;
113
114 rcu_read_unlock();
115 }
116 EXPORT_SYMBOL_GPL(disk_part_iter_init);
117
118 /**
119 * disk_part_iter_next - proceed iterator to the next partition and return it
120 * @piter: iterator of interest
121 *
122 * Proceed @piter to the next partition and return it.
123 *
124 * CONTEXT:
125 * Don't care.
126 */
disk_part_iter_next(struct disk_part_iter * piter)127 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
128 {
129 struct disk_part_tbl *ptbl;
130 int inc, end;
131
132 /* put the last partition */
133 disk_put_part(piter->part);
134 piter->part = NULL;
135
136 /* get part_tbl */
137 rcu_read_lock();
138 ptbl = rcu_dereference(piter->disk->part_tbl);
139
140 /* determine iteration parameters */
141 if (piter->flags & DISK_PITER_REVERSE) {
142 inc = -1;
143 if (piter->flags & (DISK_PITER_INCL_PART0 |
144 DISK_PITER_INCL_EMPTY_PART0))
145 end = -1;
146 else
147 end = 0;
148 } else {
149 inc = 1;
150 end = ptbl->len;
151 }
152
153 /* iterate to the next partition */
154 for (; piter->idx != end; piter->idx += inc) {
155 struct hd_struct *part;
156
157 part = rcu_dereference(ptbl->part[piter->idx]);
158 if (!part)
159 continue;
160 if (!part_nr_sects_read(part) &&
161 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
162 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
163 piter->idx == 0))
164 continue;
165
166 get_device(part_to_dev(part));
167 piter->part = part;
168 piter->idx += inc;
169 break;
170 }
171
172 rcu_read_unlock();
173
174 return piter->part;
175 }
176 EXPORT_SYMBOL_GPL(disk_part_iter_next);
177
178 /**
179 * disk_part_iter_exit - finish up partition iteration
180 * @piter: iter of interest
181 *
182 * Called when iteration is over. Cleans up @piter.
183 *
184 * CONTEXT:
185 * Don't care.
186 */
disk_part_iter_exit(struct disk_part_iter * piter)187 void disk_part_iter_exit(struct disk_part_iter *piter)
188 {
189 disk_put_part(piter->part);
190 piter->part = NULL;
191 }
192 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
193
sector_in_part(struct hd_struct * part,sector_t sector)194 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
195 {
196 return part->start_sect <= sector &&
197 sector < part->start_sect + part_nr_sects_read(part);
198 }
199
200 /**
201 * disk_map_sector_rcu - map sector to partition
202 * @disk: gendisk of interest
203 * @sector: sector to map
204 *
205 * Find out which partition @sector maps to on @disk. This is
206 * primarily used for stats accounting.
207 *
208 * CONTEXT:
209 * RCU read locked. The returned partition pointer is valid only
210 * while preemption is disabled.
211 *
212 * RETURNS:
213 * Found partition on success, part0 is returned if no partition matches
214 */
disk_map_sector_rcu(struct gendisk * disk,sector_t sector)215 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
216 {
217 struct disk_part_tbl *ptbl;
218 struct hd_struct *part;
219 int i;
220
221 ptbl = rcu_dereference(disk->part_tbl);
222
223 part = rcu_dereference(ptbl->last_lookup);
224 if (part && sector_in_part(part, sector))
225 return part;
226
227 for (i = 1; i < ptbl->len; i++) {
228 part = rcu_dereference(ptbl->part[i]);
229
230 if (part && sector_in_part(part, sector)) {
231 rcu_assign_pointer(ptbl->last_lookup, part);
232 return part;
233 }
234 }
235 return &disk->part0;
236 }
237 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
238
239 /*
240 * Can be deleted altogether. Later.
241 *
242 */
243 static struct blk_major_name {
244 struct blk_major_name *next;
245 int major;
246 char name[16];
247 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
248
249 /* index in the above - for now: assume no multimajor ranges */
major_to_index(unsigned major)250 static inline int major_to_index(unsigned major)
251 {
252 return major % BLKDEV_MAJOR_HASH_SIZE;
253 }
254
255 #ifdef CONFIG_PROC_FS
blkdev_show(struct seq_file * seqf,off_t offset)256 void blkdev_show(struct seq_file *seqf, off_t offset)
257 {
258 struct blk_major_name *dp;
259
260 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
261 mutex_lock(&block_class_lock);
262 for (dp = major_names[offset]; dp; dp = dp->next)
263 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
264 mutex_unlock(&block_class_lock);
265 }
266 }
267 #endif /* CONFIG_PROC_FS */
268
269 /**
270 * register_blkdev - register a new block device
271 *
272 * @major: the requested major device number [1..255]. If @major=0, try to
273 * allocate any unused major number.
274 * @name: the name of the new block device as a zero terminated string
275 *
276 * The @name must be unique within the system.
277 *
278 * The return value depends on the @major input parameter.
279 * - if a major device number was requested in range [1..255] then the
280 * function returns zero on success, or a negative error code
281 * - if any unused major number was requested with @major=0 parameter
282 * then the return value is the allocated major number in range
283 * [1..255] or a negative error code otherwise
284 */
register_blkdev(unsigned int major,const char * name)285 int register_blkdev(unsigned int major, const char *name)
286 {
287 struct blk_major_name **n, *p;
288 int index, ret = 0;
289
290 mutex_lock(&block_class_lock);
291
292 /* temporary */
293 if (major == 0) {
294 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
295 if (major_names[index] == NULL)
296 break;
297 }
298
299 if (index == 0) {
300 printk("register_blkdev: failed to get major for %s\n",
301 name);
302 ret = -EBUSY;
303 goto out;
304 }
305 major = index;
306 ret = major;
307 }
308
309 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
310 if (p == NULL) {
311 ret = -ENOMEM;
312 goto out;
313 }
314
315 p->major = major;
316 strlcpy(p->name, name, sizeof(p->name));
317 p->next = NULL;
318 index = major_to_index(major);
319
320 for (n = &major_names[index]; *n; n = &(*n)->next) {
321 if ((*n)->major == major)
322 break;
323 }
324 if (!*n)
325 *n = p;
326 else
327 ret = -EBUSY;
328
329 if (ret < 0) {
330 printk("register_blkdev: cannot get major %d for %s\n",
331 major, name);
332 kfree(p);
333 }
334 out:
335 mutex_unlock(&block_class_lock);
336 return ret;
337 }
338
339 EXPORT_SYMBOL(register_blkdev);
340
unregister_blkdev(unsigned int major,const char * name)341 void unregister_blkdev(unsigned int major, const char *name)
342 {
343 struct blk_major_name **n;
344 struct blk_major_name *p = NULL;
345 int index = major_to_index(major);
346
347 mutex_lock(&block_class_lock);
348 for (n = &major_names[index]; *n; n = &(*n)->next)
349 if ((*n)->major == major)
350 break;
351 if (!*n || strcmp((*n)->name, name)) {
352 WARN_ON(1);
353 } else {
354 p = *n;
355 *n = p->next;
356 }
357 mutex_unlock(&block_class_lock);
358 kfree(p);
359 }
360
361 EXPORT_SYMBOL(unregister_blkdev);
362
363 static struct kobj_map *bdev_map;
364
365 /**
366 * blk_mangle_minor - scatter minor numbers apart
367 * @minor: minor number to mangle
368 *
369 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
370 * is enabled. Mangling twice gives the original value.
371 *
372 * RETURNS:
373 * Mangled value.
374 *
375 * CONTEXT:
376 * Don't care.
377 */
blk_mangle_minor(int minor)378 static int blk_mangle_minor(int minor)
379 {
380 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
381 int i;
382
383 for (i = 0; i < MINORBITS / 2; i++) {
384 int low = minor & (1 << i);
385 int high = minor & (1 << (MINORBITS - 1 - i));
386 int distance = MINORBITS - 1 - 2 * i;
387
388 minor ^= low | high; /* clear both bits */
389 low <<= distance; /* swap the positions */
390 high >>= distance;
391 minor |= low | high; /* and set */
392 }
393 #endif
394 return minor;
395 }
396
397 /**
398 * blk_alloc_devt - allocate a dev_t for a partition
399 * @part: partition to allocate dev_t for
400 * @devt: out parameter for resulting dev_t
401 *
402 * Allocate a dev_t for block device.
403 *
404 * RETURNS:
405 * 0 on success, allocated dev_t is returned in *@devt. -errno on
406 * failure.
407 *
408 * CONTEXT:
409 * Might sleep.
410 */
blk_alloc_devt(struct hd_struct * part,dev_t * devt)411 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
412 {
413 struct gendisk *disk = part_to_disk(part);
414 int idx;
415
416 /* in consecutive minor range? */
417 if (part->partno < disk->minors) {
418 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
419 return 0;
420 }
421
422 /* allocate ext devt */
423 idr_preload(GFP_KERNEL);
424
425 spin_lock_bh(&ext_devt_lock);
426 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
427 spin_unlock_bh(&ext_devt_lock);
428
429 idr_preload_end();
430 if (idx < 0)
431 return idx == -ENOSPC ? -EBUSY : idx;
432
433 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
434 return 0;
435 }
436
437 /**
438 * blk_free_devt - free a dev_t
439 * @devt: dev_t to free
440 *
441 * Free @devt which was allocated using blk_alloc_devt().
442 *
443 * CONTEXT:
444 * Might sleep.
445 */
blk_free_devt(dev_t devt)446 void blk_free_devt(dev_t devt)
447 {
448 if (devt == MKDEV(0, 0))
449 return;
450
451 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
452 spin_lock_bh(&ext_devt_lock);
453 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
454 spin_unlock_bh(&ext_devt_lock);
455 }
456 }
457
bdevt_str(dev_t devt,char * buf)458 static char *bdevt_str(dev_t devt, char *buf)
459 {
460 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
461 char tbuf[BDEVT_SIZE];
462 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
463 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
464 } else
465 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
466
467 return buf;
468 }
469
470 /*
471 * Register device numbers dev..(dev+range-1)
472 * range must be nonzero
473 * The hash chain is sorted on range, so that subranges can override.
474 */
blk_register_region(dev_t devt,unsigned long range,struct module * module,struct kobject * (* probe)(dev_t,int *,void *),int (* lock)(dev_t,void *),void * data)475 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
476 struct kobject *(*probe)(dev_t, int *, void *),
477 int (*lock)(dev_t, void *), void *data)
478 {
479 kobj_map(bdev_map, devt, range, module, probe, lock, data);
480 }
481
482 EXPORT_SYMBOL(blk_register_region);
483
blk_unregister_region(dev_t devt,unsigned long range)484 void blk_unregister_region(dev_t devt, unsigned long range)
485 {
486 kobj_unmap(bdev_map, devt, range);
487 }
488
489 EXPORT_SYMBOL(blk_unregister_region);
490
exact_match(dev_t devt,int * partno,void * data)491 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
492 {
493 struct gendisk *p = data;
494
495 return &disk_to_dev(p)->kobj;
496 }
497
exact_lock(dev_t devt,void * data)498 static int exact_lock(dev_t devt, void *data)
499 {
500 struct gendisk *p = data;
501
502 if (!get_disk(p))
503 return -1;
504 return 0;
505 }
506
register_disk(struct gendisk * disk)507 static void register_disk(struct gendisk *disk)
508 {
509 struct device *ddev = disk_to_dev(disk);
510 struct block_device *bdev;
511 struct disk_part_iter piter;
512 struct hd_struct *part;
513 int err;
514
515 ddev->parent = disk->driverfs_dev;
516
517 dev_set_name(ddev, "%s", disk->disk_name);
518
519 /* delay uevents, until we scanned partition table */
520 dev_set_uevent_suppress(ddev, 1);
521
522 if (device_add(ddev))
523 return;
524 if (!sysfs_deprecated) {
525 err = sysfs_create_link(block_depr, &ddev->kobj,
526 kobject_name(&ddev->kobj));
527 if (err) {
528 device_del(ddev);
529 return;
530 }
531 }
532
533 /*
534 * avoid probable deadlock caused by allocating memory with
535 * GFP_KERNEL in runtime_resume callback of its all ancestor
536 * devices
537 */
538 pm_runtime_set_memalloc_noio(ddev, true);
539
540 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
541 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
542
543 /* No minors to use for partitions */
544 if (!disk_part_scan_enabled(disk))
545 goto exit;
546
547 /* No such device (e.g., media were just removed) */
548 if (!get_capacity(disk))
549 goto exit;
550
551 bdev = bdget_disk(disk, 0);
552 if (!bdev)
553 goto exit;
554
555 bdev->bd_invalidated = 1;
556 err = blkdev_get(bdev, FMODE_READ, NULL);
557 if (err < 0)
558 goto exit;
559 blkdev_put(bdev, FMODE_READ);
560
561 exit:
562 /* announce disk after possible partitions are created */
563 dev_set_uevent_suppress(ddev, 0);
564 kobject_uevent(&ddev->kobj, KOBJ_ADD);
565
566 /* announce possible partitions */
567 disk_part_iter_init(&piter, disk, 0);
568 while ((part = disk_part_iter_next(&piter)))
569 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
570 disk_part_iter_exit(&piter);
571 }
572
573 /**
574 * add_disk - add partitioning information to kernel list
575 * @disk: per-device partitioning information
576 *
577 * This function registers the partitioning information in @disk
578 * with the kernel.
579 *
580 * FIXME: error handling
581 */
add_disk(struct gendisk * disk)582 void add_disk(struct gendisk *disk)
583 {
584 struct backing_dev_info *bdi;
585 dev_t devt;
586 int retval;
587
588 /* minors == 0 indicates to use ext devt from part0 and should
589 * be accompanied with EXT_DEVT flag. Make sure all
590 * parameters make sense.
591 */
592 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
593 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
594
595 disk->flags |= GENHD_FL_UP;
596
597 retval = blk_alloc_devt(&disk->part0, &devt);
598 if (retval) {
599 WARN_ON(1);
600 return;
601 }
602 disk_to_dev(disk)->devt = devt;
603
604 /* ->major and ->first_minor aren't supposed to be
605 * dereferenced from here on, but set them just in case.
606 */
607 disk->major = MAJOR(devt);
608 disk->first_minor = MINOR(devt);
609
610 disk_alloc_events(disk);
611
612 /* Register BDI before referencing it from bdev */
613 bdi = &disk->queue->backing_dev_info;
614 bdi_register_dev(bdi, disk_devt(disk));
615
616 blk_register_region(disk_devt(disk), disk->minors, NULL,
617 exact_match, exact_lock, disk);
618 register_disk(disk);
619 blk_register_queue(disk);
620
621 /*
622 * Take an extra ref on queue which will be put on disk_release()
623 * so that it sticks around as long as @disk is there.
624 */
625 WARN_ON_ONCE(!blk_get_queue(disk->queue));
626
627 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
628 "bdi");
629 WARN_ON(retval);
630
631 disk_add_events(disk);
632 }
633 EXPORT_SYMBOL(add_disk);
634
del_gendisk(struct gendisk * disk)635 void del_gendisk(struct gendisk *disk)
636 {
637 struct disk_part_iter piter;
638 struct hd_struct *part;
639
640 disk_del_events(disk);
641
642 /* invalidate stuff */
643 disk_part_iter_init(&piter, disk,
644 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
645 while ((part = disk_part_iter_next(&piter))) {
646 invalidate_partition(disk, part->partno);
647 delete_partition(disk, part->partno);
648 }
649 disk_part_iter_exit(&piter);
650
651 invalidate_partition(disk, 0);
652 set_capacity(disk, 0);
653 disk->flags &= ~GENHD_FL_UP;
654
655 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
656 bdi_unregister(&disk->queue->backing_dev_info);
657 blk_unregister_queue(disk);
658 blk_unregister_region(disk_devt(disk), disk->minors);
659
660 part_stat_set_all(&disk->part0, 0);
661 disk->part0.stamp = 0;
662
663 kobject_put(disk->part0.holder_dir);
664 kobject_put(disk->slave_dir);
665 disk->driverfs_dev = NULL;
666 if (!sysfs_deprecated)
667 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
668 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
669 device_del(disk_to_dev(disk));
670 }
671 EXPORT_SYMBOL(del_gendisk);
672
673 /**
674 * get_gendisk - get partitioning information for a given device
675 * @devt: device to get partitioning information for
676 * @partno: returned partition index
677 *
678 * This function gets the structure containing partitioning
679 * information for the given device @devt.
680 */
get_gendisk(dev_t devt,int * partno)681 struct gendisk *get_gendisk(dev_t devt, int *partno)
682 {
683 struct gendisk *disk = NULL;
684
685 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
686 struct kobject *kobj;
687
688 kobj = kobj_lookup(bdev_map, devt, partno);
689 if (kobj)
690 disk = dev_to_disk(kobj_to_dev(kobj));
691 } else {
692 struct hd_struct *part;
693
694 spin_lock_bh(&ext_devt_lock);
695 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
696 if (part && get_disk(part_to_disk(part))) {
697 *partno = part->partno;
698 disk = part_to_disk(part);
699 }
700 spin_unlock_bh(&ext_devt_lock);
701 }
702
703 return disk;
704 }
705 EXPORT_SYMBOL(get_gendisk);
706
707 /**
708 * bdget_disk - do bdget() by gendisk and partition number
709 * @disk: gendisk of interest
710 * @partno: partition number
711 *
712 * Find partition @partno from @disk, do bdget() on it.
713 *
714 * CONTEXT:
715 * Don't care.
716 *
717 * RETURNS:
718 * Resulting block_device on success, NULL on failure.
719 */
bdget_disk(struct gendisk * disk,int partno)720 struct block_device *bdget_disk(struct gendisk *disk, int partno)
721 {
722 struct hd_struct *part;
723 struct block_device *bdev = NULL;
724
725 part = disk_get_part(disk, partno);
726 if (part)
727 bdev = bdget(part_devt(part));
728 disk_put_part(part);
729
730 return bdev;
731 }
732 EXPORT_SYMBOL(bdget_disk);
733
734 /*
735 * print a full list of all partitions - intended for places where the root
736 * filesystem can't be mounted and thus to give the victim some idea of what
737 * went wrong
738 */
printk_all_partitions(void)739 void __init printk_all_partitions(void)
740 {
741 struct class_dev_iter iter;
742 struct device *dev;
743
744 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
745 while ((dev = class_dev_iter_next(&iter))) {
746 struct gendisk *disk = dev_to_disk(dev);
747 struct disk_part_iter piter;
748 struct hd_struct *part;
749 char name_buf[BDEVNAME_SIZE];
750 char devt_buf[BDEVT_SIZE];
751
752 /*
753 * Don't show empty devices or things that have been
754 * suppressed
755 */
756 if (get_capacity(disk) == 0 ||
757 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
758 continue;
759
760 /*
761 * Note, unlike /proc/partitions, I am showing the
762 * numbers in hex - the same format as the root=
763 * option takes.
764 */
765 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
766 while ((part = disk_part_iter_next(&piter))) {
767 bool is_part0 = part == &disk->part0;
768
769 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
770 bdevt_str(part_devt(part), devt_buf),
771 (unsigned long long)part_nr_sects_read(part) >> 1
772 , disk_name(disk, part->partno, name_buf),
773 part->info ? part->info->uuid : "");
774 if (is_part0) {
775 if (disk->driverfs_dev != NULL &&
776 disk->driverfs_dev->driver != NULL)
777 printk(" driver: %s\n",
778 disk->driverfs_dev->driver->name);
779 else
780 printk(" (driver?)\n");
781 } else
782 printk("\n");
783 }
784 disk_part_iter_exit(&piter);
785 }
786 class_dev_iter_exit(&iter);
787 }
788
789 #ifdef CONFIG_PROC_FS
790 /* iterator */
disk_seqf_start(struct seq_file * seqf,loff_t * pos)791 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
792 {
793 loff_t skip = *pos;
794 struct class_dev_iter *iter;
795 struct device *dev;
796
797 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
798 if (!iter)
799 return ERR_PTR(-ENOMEM);
800
801 seqf->private = iter;
802 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
803 do {
804 dev = class_dev_iter_next(iter);
805 if (!dev)
806 return NULL;
807 } while (skip--);
808
809 return dev_to_disk(dev);
810 }
811
disk_seqf_next(struct seq_file * seqf,void * v,loff_t * pos)812 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
813 {
814 struct device *dev;
815
816 (*pos)++;
817 dev = class_dev_iter_next(seqf->private);
818 if (dev)
819 return dev_to_disk(dev);
820
821 return NULL;
822 }
823
disk_seqf_stop(struct seq_file * seqf,void * v)824 static void disk_seqf_stop(struct seq_file *seqf, void *v)
825 {
826 struct class_dev_iter *iter = seqf->private;
827
828 /* stop is called even after start failed :-( */
829 if (iter) {
830 class_dev_iter_exit(iter);
831 kfree(iter);
832 seqf->private = NULL;
833 }
834 }
835
show_partition_start(struct seq_file * seqf,loff_t * pos)836 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
837 {
838 void *p;
839
840 p = disk_seqf_start(seqf, pos);
841 if (!IS_ERR_OR_NULL(p) && !*pos)
842 seq_puts(seqf, "major minor #blocks name\n\n");
843 return p;
844 }
845
show_partition(struct seq_file * seqf,void * v)846 static int show_partition(struct seq_file *seqf, void *v)
847 {
848 struct gendisk *sgp = v;
849 struct disk_part_iter piter;
850 struct hd_struct *part;
851 char buf[BDEVNAME_SIZE];
852
853 /* Don't show non-partitionable removeable devices or empty devices */
854 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
855 (sgp->flags & GENHD_FL_REMOVABLE)))
856 return 0;
857 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
858 return 0;
859
860 /* show the full disk and all non-0 size partitions of it */
861 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
862 while ((part = disk_part_iter_next(&piter)))
863 seq_printf(seqf, "%4d %7d %10llu %s\n",
864 MAJOR(part_devt(part)), MINOR(part_devt(part)),
865 (unsigned long long)part_nr_sects_read(part) >> 1,
866 disk_name(sgp, part->partno, buf));
867 disk_part_iter_exit(&piter);
868
869 return 0;
870 }
871
872 static const struct seq_operations partitions_op = {
873 .start = show_partition_start,
874 .next = disk_seqf_next,
875 .stop = disk_seqf_stop,
876 .show = show_partition
877 };
878
partitions_open(struct inode * inode,struct file * file)879 static int partitions_open(struct inode *inode, struct file *file)
880 {
881 return seq_open(file, &partitions_op);
882 }
883
884 static const struct file_operations proc_partitions_operations = {
885 .open = partitions_open,
886 .read = seq_read,
887 .llseek = seq_lseek,
888 .release = seq_release,
889 };
890 #endif
891
892
base_probe(dev_t devt,int * partno,void * data)893 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
894 {
895 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
896 /* Make old-style 2.4 aliases work */
897 request_module("block-major-%d", MAJOR(devt));
898 return NULL;
899 }
900
genhd_device_init(void)901 static int __init genhd_device_init(void)
902 {
903 int error;
904
905 block_class.dev_kobj = sysfs_dev_block_kobj;
906 error = class_register(&block_class);
907 if (unlikely(error))
908 return error;
909 bdev_map = kobj_map_init(base_probe, &block_class_lock);
910 blk_dev_init();
911
912 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
913
914 /* create top-level block dir */
915 if (!sysfs_deprecated)
916 block_depr = kobject_create_and_add("block", NULL);
917 return 0;
918 }
919
920 subsys_initcall(genhd_device_init);
921
disk_range_show(struct device * dev,struct device_attribute * attr,char * buf)922 static ssize_t disk_range_show(struct device *dev,
923 struct device_attribute *attr, char *buf)
924 {
925 struct gendisk *disk = dev_to_disk(dev);
926
927 return sprintf(buf, "%d\n", disk->minors);
928 }
929
disk_ext_range_show(struct device * dev,struct device_attribute * attr,char * buf)930 static ssize_t disk_ext_range_show(struct device *dev,
931 struct device_attribute *attr, char *buf)
932 {
933 struct gendisk *disk = dev_to_disk(dev);
934
935 return sprintf(buf, "%d\n", disk_max_parts(disk));
936 }
937
disk_removable_show(struct device * dev,struct device_attribute * attr,char * buf)938 static ssize_t disk_removable_show(struct device *dev,
939 struct device_attribute *attr, char *buf)
940 {
941 struct gendisk *disk = dev_to_disk(dev);
942
943 return sprintf(buf, "%d\n",
944 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
945 }
946
disk_ro_show(struct device * dev,struct device_attribute * attr,char * buf)947 static ssize_t disk_ro_show(struct device *dev,
948 struct device_attribute *attr, char *buf)
949 {
950 struct gendisk *disk = dev_to_disk(dev);
951
952 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
953 }
954
disk_capability_show(struct device * dev,struct device_attribute * attr,char * buf)955 static ssize_t disk_capability_show(struct device *dev,
956 struct device_attribute *attr, char *buf)
957 {
958 struct gendisk *disk = dev_to_disk(dev);
959
960 return sprintf(buf, "%x\n", disk->flags);
961 }
962
disk_alignment_offset_show(struct device * dev,struct device_attribute * attr,char * buf)963 static ssize_t disk_alignment_offset_show(struct device *dev,
964 struct device_attribute *attr,
965 char *buf)
966 {
967 struct gendisk *disk = dev_to_disk(dev);
968
969 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
970 }
971
disk_discard_alignment_show(struct device * dev,struct device_attribute * attr,char * buf)972 static ssize_t disk_discard_alignment_show(struct device *dev,
973 struct device_attribute *attr,
974 char *buf)
975 {
976 struct gendisk *disk = dev_to_disk(dev);
977
978 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
979 }
980
981 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
982 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
983 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
984 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
985 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
986 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
987 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
988 NULL);
989 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
990 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
991 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
992 #ifdef CONFIG_FAIL_MAKE_REQUEST
993 static struct device_attribute dev_attr_fail =
994 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
995 #endif
996 #ifdef CONFIG_FAIL_IO_TIMEOUT
997 static struct device_attribute dev_attr_fail_timeout =
998 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
999 part_timeout_store);
1000 #endif
1001
1002 static struct attribute *disk_attrs[] = {
1003 &dev_attr_range.attr,
1004 &dev_attr_ext_range.attr,
1005 &dev_attr_removable.attr,
1006 &dev_attr_ro.attr,
1007 &dev_attr_size.attr,
1008 &dev_attr_alignment_offset.attr,
1009 &dev_attr_discard_alignment.attr,
1010 &dev_attr_capability.attr,
1011 &dev_attr_stat.attr,
1012 &dev_attr_inflight.attr,
1013 #ifdef CONFIG_FAIL_MAKE_REQUEST
1014 &dev_attr_fail.attr,
1015 #endif
1016 #ifdef CONFIG_FAIL_IO_TIMEOUT
1017 &dev_attr_fail_timeout.attr,
1018 #endif
1019 NULL
1020 };
1021
1022 static struct attribute_group disk_attr_group = {
1023 .attrs = disk_attrs,
1024 };
1025
1026 static const struct attribute_group *disk_attr_groups[] = {
1027 &disk_attr_group,
1028 NULL
1029 };
1030
1031 /**
1032 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1033 * @disk: disk to replace part_tbl for
1034 * @new_ptbl: new part_tbl to install
1035 *
1036 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1037 * original ptbl is freed using RCU callback.
1038 *
1039 * LOCKING:
1040 * Matching bd_mutx locked.
1041 */
disk_replace_part_tbl(struct gendisk * disk,struct disk_part_tbl * new_ptbl)1042 static void disk_replace_part_tbl(struct gendisk *disk,
1043 struct disk_part_tbl *new_ptbl)
1044 {
1045 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1046
1047 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1048
1049 if (old_ptbl) {
1050 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1051 kfree_rcu(old_ptbl, rcu_head);
1052 }
1053 }
1054
1055 /**
1056 * disk_expand_part_tbl - expand disk->part_tbl
1057 * @disk: disk to expand part_tbl for
1058 * @partno: expand such that this partno can fit in
1059 *
1060 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1061 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1062 *
1063 * LOCKING:
1064 * Matching bd_mutex locked, might sleep.
1065 *
1066 * RETURNS:
1067 * 0 on success, -errno on failure.
1068 */
disk_expand_part_tbl(struct gendisk * disk,int partno)1069 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1070 {
1071 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1072 struct disk_part_tbl *new_ptbl;
1073 int len = old_ptbl ? old_ptbl->len : 0;
1074 int i, target;
1075 size_t size;
1076
1077 /*
1078 * check for int overflow, since we can get here from blkpg_ioctl()
1079 * with a user passed 'partno'.
1080 */
1081 target = partno + 1;
1082 if (target < 0)
1083 return -EINVAL;
1084
1085 /* disk_max_parts() is zero during initialization, ignore if so */
1086 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1087 return -EINVAL;
1088
1089 if (target <= len)
1090 return 0;
1091
1092 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1093 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1094 if (!new_ptbl)
1095 return -ENOMEM;
1096
1097 new_ptbl->len = target;
1098
1099 for (i = 0; i < len; i++)
1100 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1101
1102 disk_replace_part_tbl(disk, new_ptbl);
1103 return 0;
1104 }
1105
disk_release(struct device * dev)1106 static void disk_release(struct device *dev)
1107 {
1108 struct gendisk *disk = dev_to_disk(dev);
1109
1110 blk_free_devt(dev->devt);
1111 disk_release_events(disk);
1112 kfree(disk->random);
1113 disk_replace_part_tbl(disk, NULL);
1114 free_part_stats(&disk->part0);
1115 free_part_info(&disk->part0);
1116 if (disk->queue)
1117 blk_put_queue(disk->queue);
1118 kfree(disk);
1119 }
1120
disk_uevent(struct device * dev,struct kobj_uevent_env * env)1121 static int disk_uevent(struct device *dev, struct kobj_uevent_env *env)
1122 {
1123 struct gendisk *disk = dev_to_disk(dev);
1124 struct disk_part_iter piter;
1125 struct hd_struct *part;
1126 int cnt = 0;
1127
1128 disk_part_iter_init(&piter, disk, 0);
1129 while((part = disk_part_iter_next(&piter)))
1130 cnt++;
1131 disk_part_iter_exit(&piter);
1132 add_uevent_var(env, "NPARTS=%u", cnt);
1133 return 0;
1134 }
1135
1136 struct class block_class = {
1137 .name = "block",
1138 };
1139
block_devnode(struct device * dev,umode_t * mode,kuid_t * uid,kgid_t * gid)1140 static char *block_devnode(struct device *dev, umode_t *mode,
1141 kuid_t *uid, kgid_t *gid)
1142 {
1143 struct gendisk *disk = dev_to_disk(dev);
1144
1145 if (disk->devnode)
1146 return disk->devnode(disk, mode);
1147 return NULL;
1148 }
1149
1150 static struct device_type disk_type = {
1151 .name = "disk",
1152 .groups = disk_attr_groups,
1153 .release = disk_release,
1154 .devnode = block_devnode,
1155 .uevent = disk_uevent,
1156 };
1157
1158 #ifdef CONFIG_PROC_FS
1159 /*
1160 * aggregate disk stat collector. Uses the same stats that the sysfs
1161 * entries do, above, but makes them available through one seq_file.
1162 *
1163 * The output looks suspiciously like /proc/partitions with a bunch of
1164 * extra fields.
1165 */
diskstats_show(struct seq_file * seqf,void * v)1166 static int diskstats_show(struct seq_file *seqf, void *v)
1167 {
1168 struct gendisk *gp = v;
1169 struct disk_part_iter piter;
1170 struct hd_struct *hd;
1171 char buf[BDEVNAME_SIZE];
1172 int cpu;
1173
1174 /*
1175 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1176 seq_puts(seqf, "major minor name"
1177 " rio rmerge rsect ruse wio wmerge "
1178 "wsect wuse running use aveq"
1179 "\n\n");
1180 */
1181
1182 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1183 while ((hd = disk_part_iter_next(&piter))) {
1184 cpu = part_stat_lock();
1185 part_round_stats(cpu, hd);
1186 part_stat_unlock();
1187 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1188 "%u %lu %lu %lu %u %u %u %u\n",
1189 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1190 disk_name(gp, hd->partno, buf),
1191 part_stat_read(hd, ios[READ]),
1192 part_stat_read(hd, merges[READ]),
1193 part_stat_read(hd, sectors[READ]),
1194 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1195 part_stat_read(hd, ios[WRITE]),
1196 part_stat_read(hd, merges[WRITE]),
1197 part_stat_read(hd, sectors[WRITE]),
1198 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1199 part_in_flight(hd),
1200 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1201 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1202 );
1203 }
1204 disk_part_iter_exit(&piter);
1205
1206 return 0;
1207 }
1208
1209 static const struct seq_operations diskstats_op = {
1210 .start = disk_seqf_start,
1211 .next = disk_seqf_next,
1212 .stop = disk_seqf_stop,
1213 .show = diskstats_show
1214 };
1215
diskstats_open(struct inode * inode,struct file * file)1216 static int diskstats_open(struct inode *inode, struct file *file)
1217 {
1218 return seq_open(file, &diskstats_op);
1219 }
1220
1221 static const struct file_operations proc_diskstats_operations = {
1222 .open = diskstats_open,
1223 .read = seq_read,
1224 .llseek = seq_lseek,
1225 .release = seq_release,
1226 };
1227
proc_genhd_init(void)1228 static int __init proc_genhd_init(void)
1229 {
1230 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1231 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1232 return 0;
1233 }
1234 module_init(proc_genhd_init);
1235 #endif /* CONFIG_PROC_FS */
1236
blk_lookup_devt(const char * name,int partno)1237 dev_t blk_lookup_devt(const char *name, int partno)
1238 {
1239 dev_t devt = MKDEV(0, 0);
1240 struct class_dev_iter iter;
1241 struct device *dev;
1242
1243 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1244 while ((dev = class_dev_iter_next(&iter))) {
1245 struct gendisk *disk = dev_to_disk(dev);
1246 struct hd_struct *part;
1247
1248 if (strcmp(dev_name(dev), name))
1249 continue;
1250
1251 if (partno < disk->minors) {
1252 /* We need to return the right devno, even
1253 * if the partition doesn't exist yet.
1254 */
1255 devt = MKDEV(MAJOR(dev->devt),
1256 MINOR(dev->devt) + partno);
1257 break;
1258 }
1259 part = disk_get_part(disk, partno);
1260 if (part) {
1261 devt = part_devt(part);
1262 disk_put_part(part);
1263 break;
1264 }
1265 disk_put_part(part);
1266 }
1267 class_dev_iter_exit(&iter);
1268 return devt;
1269 }
1270 EXPORT_SYMBOL(blk_lookup_devt);
1271
alloc_disk(int minors)1272 struct gendisk *alloc_disk(int minors)
1273 {
1274 return alloc_disk_node(minors, NUMA_NO_NODE);
1275 }
1276 EXPORT_SYMBOL(alloc_disk);
1277
alloc_disk_node(int minors,int node_id)1278 struct gendisk *alloc_disk_node(int minors, int node_id)
1279 {
1280 struct gendisk *disk;
1281
1282 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1283 if (disk) {
1284 if (!init_part_stats(&disk->part0)) {
1285 kfree(disk);
1286 return NULL;
1287 }
1288 disk->node_id = node_id;
1289 if (disk_expand_part_tbl(disk, 0)) {
1290 free_part_stats(&disk->part0);
1291 kfree(disk);
1292 return NULL;
1293 }
1294 disk->part_tbl->part[0] = &disk->part0;
1295
1296 /*
1297 * set_capacity() and get_capacity() currently don't use
1298 * seqcounter to read/update the part0->nr_sects. Still init
1299 * the counter as we can read the sectors in IO submission
1300 * patch using seqence counters.
1301 *
1302 * TODO: Ideally set_capacity() and get_capacity() should be
1303 * converted to make use of bd_mutex and sequence counters.
1304 */
1305 seqcount_init(&disk->part0.nr_sects_seq);
1306 hd_ref_init(&disk->part0);
1307
1308 disk->minors = minors;
1309 rand_initialize_disk(disk);
1310 disk_to_dev(disk)->class = &block_class;
1311 disk_to_dev(disk)->type = &disk_type;
1312 device_initialize(disk_to_dev(disk));
1313 }
1314 return disk;
1315 }
1316 EXPORT_SYMBOL(alloc_disk_node);
1317
get_disk(struct gendisk * disk)1318 struct kobject *get_disk(struct gendisk *disk)
1319 {
1320 struct module *owner;
1321 struct kobject *kobj;
1322
1323 if (!disk->fops)
1324 return NULL;
1325 owner = disk->fops->owner;
1326 if (owner && !try_module_get(owner))
1327 return NULL;
1328 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1329 if (kobj == NULL) {
1330 module_put(owner);
1331 return NULL;
1332 }
1333 return kobj;
1334
1335 }
1336
1337 EXPORT_SYMBOL(get_disk);
1338
put_disk(struct gendisk * disk)1339 void put_disk(struct gendisk *disk)
1340 {
1341 if (disk)
1342 kobject_put(&disk_to_dev(disk)->kobj);
1343 }
1344
1345 EXPORT_SYMBOL(put_disk);
1346
set_disk_ro_uevent(struct gendisk * gd,int ro)1347 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1348 {
1349 char event[] = "DISK_RO=1";
1350 char *envp[] = { event, NULL };
1351
1352 if (!ro)
1353 event[8] = '0';
1354 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1355 }
1356
set_device_ro(struct block_device * bdev,int flag)1357 void set_device_ro(struct block_device *bdev, int flag)
1358 {
1359 bdev->bd_part->policy = flag;
1360 }
1361
1362 EXPORT_SYMBOL(set_device_ro);
1363
set_disk_ro(struct gendisk * disk,int flag)1364 void set_disk_ro(struct gendisk *disk, int flag)
1365 {
1366 struct disk_part_iter piter;
1367 struct hd_struct *part;
1368
1369 if (disk->part0.policy != flag) {
1370 set_disk_ro_uevent(disk, flag);
1371 disk->part0.policy = flag;
1372 }
1373
1374 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1375 while ((part = disk_part_iter_next(&piter)))
1376 part->policy = flag;
1377 disk_part_iter_exit(&piter);
1378 }
1379
1380 EXPORT_SYMBOL(set_disk_ro);
1381
bdev_read_only(struct block_device * bdev)1382 int bdev_read_only(struct block_device *bdev)
1383 {
1384 if (!bdev)
1385 return 0;
1386 return bdev->bd_part->policy;
1387 }
1388
1389 EXPORT_SYMBOL(bdev_read_only);
1390
invalidate_partition(struct gendisk * disk,int partno)1391 int invalidate_partition(struct gendisk *disk, int partno)
1392 {
1393 int res = 0;
1394 struct block_device *bdev = bdget_disk(disk, partno);
1395 if (bdev) {
1396 fsync_bdev(bdev);
1397 res = __invalidate_device(bdev, true);
1398 bdput(bdev);
1399 }
1400 return res;
1401 }
1402
1403 EXPORT_SYMBOL(invalidate_partition);
1404
1405 /*
1406 * Disk events - monitor disk events like media change and eject request.
1407 */
1408 struct disk_events {
1409 struct list_head node; /* all disk_event's */
1410 struct gendisk *disk; /* the associated disk */
1411 spinlock_t lock;
1412
1413 struct mutex block_mutex; /* protects blocking */
1414 int block; /* event blocking depth */
1415 unsigned int pending; /* events already sent out */
1416 unsigned int clearing; /* events being cleared */
1417
1418 long poll_msecs; /* interval, -1 for default */
1419 struct delayed_work dwork;
1420 };
1421
1422 static const char *disk_events_strs[] = {
1423 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1424 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1425 };
1426
1427 static char *disk_uevents[] = {
1428 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1429 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1430 };
1431
1432 /* list of all disk_events */
1433 static DEFINE_MUTEX(disk_events_mutex);
1434 static LIST_HEAD(disk_events);
1435
1436 /* disable in-kernel polling by default */
1437 static unsigned long disk_events_dfl_poll_msecs = 0;
1438
disk_events_poll_jiffies(struct gendisk * disk)1439 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1440 {
1441 struct disk_events *ev = disk->ev;
1442 long intv_msecs = 0;
1443
1444 /*
1445 * If device-specific poll interval is set, always use it. If
1446 * the default is being used, poll iff there are events which
1447 * can't be monitored asynchronously.
1448 */
1449 if (ev->poll_msecs >= 0)
1450 intv_msecs = ev->poll_msecs;
1451 else if (disk->events & ~disk->async_events)
1452 intv_msecs = disk_events_dfl_poll_msecs;
1453
1454 return msecs_to_jiffies(intv_msecs);
1455 }
1456
1457 /**
1458 * disk_block_events - block and flush disk event checking
1459 * @disk: disk to block events for
1460 *
1461 * On return from this function, it is guaranteed that event checking
1462 * isn't in progress and won't happen until unblocked by
1463 * disk_unblock_events(). Events blocking is counted and the actual
1464 * unblocking happens after the matching number of unblocks are done.
1465 *
1466 * Note that this intentionally does not block event checking from
1467 * disk_clear_events().
1468 *
1469 * CONTEXT:
1470 * Might sleep.
1471 */
disk_block_events(struct gendisk * disk)1472 void disk_block_events(struct gendisk *disk)
1473 {
1474 struct disk_events *ev = disk->ev;
1475 unsigned long flags;
1476 bool cancel;
1477
1478 if (!ev)
1479 return;
1480
1481 /*
1482 * Outer mutex ensures that the first blocker completes canceling
1483 * the event work before further blockers are allowed to finish.
1484 */
1485 mutex_lock(&ev->block_mutex);
1486
1487 spin_lock_irqsave(&ev->lock, flags);
1488 cancel = !ev->block++;
1489 spin_unlock_irqrestore(&ev->lock, flags);
1490
1491 if (cancel)
1492 cancel_delayed_work_sync(&disk->ev->dwork);
1493
1494 mutex_unlock(&ev->block_mutex);
1495 }
1496
__disk_unblock_events(struct gendisk * disk,bool check_now)1497 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1498 {
1499 struct disk_events *ev = disk->ev;
1500 unsigned long intv;
1501 unsigned long flags;
1502
1503 spin_lock_irqsave(&ev->lock, flags);
1504
1505 if (WARN_ON_ONCE(ev->block <= 0))
1506 goto out_unlock;
1507
1508 if (--ev->block)
1509 goto out_unlock;
1510
1511 /*
1512 * Not exactly a latency critical operation, set poll timer
1513 * slack to 25% and kick event check.
1514 */
1515 intv = disk_events_poll_jiffies(disk);
1516 set_timer_slack(&ev->dwork.timer, intv / 4);
1517 if (check_now)
1518 queue_delayed_work(system_freezable_power_efficient_wq,
1519 &ev->dwork, 0);
1520 else if (intv)
1521 queue_delayed_work(system_freezable_power_efficient_wq,
1522 &ev->dwork, intv);
1523 out_unlock:
1524 spin_unlock_irqrestore(&ev->lock, flags);
1525 }
1526
1527 /**
1528 * disk_unblock_events - unblock disk event checking
1529 * @disk: disk to unblock events for
1530 *
1531 * Undo disk_block_events(). When the block count reaches zero, it
1532 * starts events polling if configured.
1533 *
1534 * CONTEXT:
1535 * Don't care. Safe to call from irq context.
1536 */
disk_unblock_events(struct gendisk * disk)1537 void disk_unblock_events(struct gendisk *disk)
1538 {
1539 if (disk->ev)
1540 __disk_unblock_events(disk, false);
1541 }
1542
1543 /**
1544 * disk_flush_events - schedule immediate event checking and flushing
1545 * @disk: disk to check and flush events for
1546 * @mask: events to flush
1547 *
1548 * Schedule immediate event checking on @disk if not blocked. Events in
1549 * @mask are scheduled to be cleared from the driver. Note that this
1550 * doesn't clear the events from @disk->ev.
1551 *
1552 * CONTEXT:
1553 * If @mask is non-zero must be called with bdev->bd_mutex held.
1554 */
disk_flush_events(struct gendisk * disk,unsigned int mask)1555 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1556 {
1557 struct disk_events *ev = disk->ev;
1558
1559 if (!ev)
1560 return;
1561
1562 spin_lock_irq(&ev->lock);
1563 ev->clearing |= mask;
1564 if (!ev->block)
1565 mod_delayed_work(system_freezable_power_efficient_wq,
1566 &ev->dwork, 0);
1567 spin_unlock_irq(&ev->lock);
1568 }
1569
1570 /**
1571 * disk_clear_events - synchronously check, clear and return pending events
1572 * @disk: disk to fetch and clear events from
1573 * @mask: mask of events to be fetched and cleared
1574 *
1575 * Disk events are synchronously checked and pending events in @mask
1576 * are cleared and returned. This ignores the block count.
1577 *
1578 * CONTEXT:
1579 * Might sleep.
1580 */
disk_clear_events(struct gendisk * disk,unsigned int mask)1581 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1582 {
1583 const struct block_device_operations *bdops = disk->fops;
1584 struct disk_events *ev = disk->ev;
1585 unsigned int pending;
1586 unsigned int clearing = mask;
1587
1588 if (!ev) {
1589 /* for drivers still using the old ->media_changed method */
1590 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1591 bdops->media_changed && bdops->media_changed(disk))
1592 return DISK_EVENT_MEDIA_CHANGE;
1593 return 0;
1594 }
1595
1596 disk_block_events(disk);
1597
1598 /*
1599 * store the union of mask and ev->clearing on the stack so that the
1600 * race with disk_flush_events does not cause ambiguity (ev->clearing
1601 * can still be modified even if events are blocked).
1602 */
1603 spin_lock_irq(&ev->lock);
1604 clearing |= ev->clearing;
1605 ev->clearing = 0;
1606 spin_unlock_irq(&ev->lock);
1607
1608 disk_check_events(ev, &clearing);
1609 /*
1610 * if ev->clearing is not 0, the disk_flush_events got called in the
1611 * middle of this function, so we want to run the workfn without delay.
1612 */
1613 __disk_unblock_events(disk, ev->clearing ? true : false);
1614
1615 /* then, fetch and clear pending events */
1616 spin_lock_irq(&ev->lock);
1617 pending = ev->pending & mask;
1618 ev->pending &= ~mask;
1619 spin_unlock_irq(&ev->lock);
1620 WARN_ON_ONCE(clearing & mask);
1621
1622 return pending;
1623 }
1624
1625 /*
1626 * Separate this part out so that a different pointer for clearing_ptr can be
1627 * passed in for disk_clear_events.
1628 */
disk_events_workfn(struct work_struct * work)1629 static void disk_events_workfn(struct work_struct *work)
1630 {
1631 struct delayed_work *dwork = to_delayed_work(work);
1632 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1633
1634 disk_check_events(ev, &ev->clearing);
1635 }
1636
disk_check_events(struct disk_events * ev,unsigned int * clearing_ptr)1637 static void disk_check_events(struct disk_events *ev,
1638 unsigned int *clearing_ptr)
1639 {
1640 struct gendisk *disk = ev->disk;
1641 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1642 unsigned int clearing = *clearing_ptr;
1643 unsigned int events;
1644 unsigned long intv;
1645 int nr_events = 0, i;
1646
1647 /* check events */
1648 events = disk->fops->check_events(disk, clearing);
1649
1650 /* accumulate pending events and schedule next poll if necessary */
1651 spin_lock_irq(&ev->lock);
1652
1653 events &= ~ev->pending;
1654 ev->pending |= events;
1655 *clearing_ptr &= ~clearing;
1656
1657 intv = disk_events_poll_jiffies(disk);
1658 if (!ev->block && intv)
1659 queue_delayed_work(system_freezable_power_efficient_wq,
1660 &ev->dwork, intv);
1661
1662 spin_unlock_irq(&ev->lock);
1663
1664 /*
1665 * Tell userland about new events. Only the events listed in
1666 * @disk->events are reported. Unlisted events are processed the
1667 * same internally but never get reported to userland.
1668 */
1669 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1670 if (events & disk->events & (1 << i))
1671 envp[nr_events++] = disk_uevents[i];
1672
1673 if (nr_events)
1674 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1675 }
1676
1677 /*
1678 * A disk events enabled device has the following sysfs nodes under
1679 * its /sys/block/X/ directory.
1680 *
1681 * events : list of all supported events
1682 * events_async : list of events which can be detected w/o polling
1683 * events_poll_msecs : polling interval, 0: disable, -1: system default
1684 */
__disk_events_show(unsigned int events,char * buf)1685 static ssize_t __disk_events_show(unsigned int events, char *buf)
1686 {
1687 const char *delim = "";
1688 ssize_t pos = 0;
1689 int i;
1690
1691 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1692 if (events & (1 << i)) {
1693 pos += sprintf(buf + pos, "%s%s",
1694 delim, disk_events_strs[i]);
1695 delim = " ";
1696 }
1697 if (pos)
1698 pos += sprintf(buf + pos, "\n");
1699 return pos;
1700 }
1701
disk_events_show(struct device * dev,struct device_attribute * attr,char * buf)1702 static ssize_t disk_events_show(struct device *dev,
1703 struct device_attribute *attr, char *buf)
1704 {
1705 struct gendisk *disk = dev_to_disk(dev);
1706
1707 return __disk_events_show(disk->events, buf);
1708 }
1709
disk_events_async_show(struct device * dev,struct device_attribute * attr,char * buf)1710 static ssize_t disk_events_async_show(struct device *dev,
1711 struct device_attribute *attr, char *buf)
1712 {
1713 struct gendisk *disk = dev_to_disk(dev);
1714
1715 return __disk_events_show(disk->async_events, buf);
1716 }
1717
disk_events_poll_msecs_show(struct device * dev,struct device_attribute * attr,char * buf)1718 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1719 struct device_attribute *attr,
1720 char *buf)
1721 {
1722 struct gendisk *disk = dev_to_disk(dev);
1723
1724 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1725 }
1726
disk_events_poll_msecs_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)1727 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1728 struct device_attribute *attr,
1729 const char *buf, size_t count)
1730 {
1731 struct gendisk *disk = dev_to_disk(dev);
1732 long intv;
1733
1734 if (!count || !sscanf(buf, "%ld", &intv))
1735 return -EINVAL;
1736
1737 if (intv < 0 && intv != -1)
1738 return -EINVAL;
1739
1740 disk_block_events(disk);
1741 disk->ev->poll_msecs = intv;
1742 __disk_unblock_events(disk, true);
1743
1744 return count;
1745 }
1746
1747 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1748 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1749 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1750 disk_events_poll_msecs_show,
1751 disk_events_poll_msecs_store);
1752
1753 static const struct attribute *disk_events_attrs[] = {
1754 &dev_attr_events.attr,
1755 &dev_attr_events_async.attr,
1756 &dev_attr_events_poll_msecs.attr,
1757 NULL,
1758 };
1759
1760 /*
1761 * The default polling interval can be specified by the kernel
1762 * parameter block.events_dfl_poll_msecs which defaults to 0
1763 * (disable). This can also be modified runtime by writing to
1764 * /sys/module/block/events_dfl_poll_msecs.
1765 */
disk_events_set_dfl_poll_msecs(const char * val,const struct kernel_param * kp)1766 static int disk_events_set_dfl_poll_msecs(const char *val,
1767 const struct kernel_param *kp)
1768 {
1769 struct disk_events *ev;
1770 int ret;
1771
1772 ret = param_set_ulong(val, kp);
1773 if (ret < 0)
1774 return ret;
1775
1776 mutex_lock(&disk_events_mutex);
1777
1778 list_for_each_entry(ev, &disk_events, node)
1779 disk_flush_events(ev->disk, 0);
1780
1781 mutex_unlock(&disk_events_mutex);
1782
1783 return 0;
1784 }
1785
1786 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1787 .set = disk_events_set_dfl_poll_msecs,
1788 .get = param_get_ulong,
1789 };
1790
1791 #undef MODULE_PARAM_PREFIX
1792 #define MODULE_PARAM_PREFIX "block."
1793
1794 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1795 &disk_events_dfl_poll_msecs, 0644);
1796
1797 /*
1798 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1799 */
disk_alloc_events(struct gendisk * disk)1800 static void disk_alloc_events(struct gendisk *disk)
1801 {
1802 struct disk_events *ev;
1803
1804 if (!disk->fops->check_events)
1805 return;
1806
1807 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1808 if (!ev) {
1809 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1810 return;
1811 }
1812
1813 INIT_LIST_HEAD(&ev->node);
1814 ev->disk = disk;
1815 spin_lock_init(&ev->lock);
1816 mutex_init(&ev->block_mutex);
1817 ev->block = 1;
1818 ev->poll_msecs = -1;
1819 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1820
1821 disk->ev = ev;
1822 }
1823
disk_add_events(struct gendisk * disk)1824 static void disk_add_events(struct gendisk *disk)
1825 {
1826 if (!disk->ev)
1827 return;
1828
1829 /* FIXME: error handling */
1830 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1831 pr_warn("%s: failed to create sysfs files for events\n",
1832 disk->disk_name);
1833
1834 mutex_lock(&disk_events_mutex);
1835 list_add_tail(&disk->ev->node, &disk_events);
1836 mutex_unlock(&disk_events_mutex);
1837
1838 /*
1839 * Block count is initialized to 1 and the following initial
1840 * unblock kicks it into action.
1841 */
1842 __disk_unblock_events(disk, true);
1843 }
1844
disk_del_events(struct gendisk * disk)1845 static void disk_del_events(struct gendisk *disk)
1846 {
1847 if (!disk->ev)
1848 return;
1849
1850 disk_block_events(disk);
1851
1852 mutex_lock(&disk_events_mutex);
1853 list_del_init(&disk->ev->node);
1854 mutex_unlock(&disk_events_mutex);
1855
1856 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1857 }
1858
disk_release_events(struct gendisk * disk)1859 static void disk_release_events(struct gendisk *disk)
1860 {
1861 /* the block count should be 1 from disk_del_events() */
1862 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1863 kfree(disk->ev);
1864 }
1865